✨ Support OpenFL framework in federated-learning-controller (open-cluster-management-io#76)

* add openfl embeded fs

Signed-off-by: mrrr61 <mrrr61@outlook.com>

* add support for OpenFL framework

Signed-off-by: mrrr61 <mrrr61@outlook.com>

* remove ustc mirror pypi

Signed-off-by: mrrr61 <mrrr61@outlook.com>

* use keras 3.11.3

Signed-off-by: mrrr61 <mrrr61@outlook.com>

* remove proxy env

Signed-off-by: mrrr61 <mrrr61@outlook.com>

* move openfl cluster claim samples to openfl examples directory

Signed-off-by: mrrr61 <mrrr61@outlook.com>

* remove unused placement generation code

Signed-off-by: mrrr61 <mrrr61@outlook.com>

---------

Signed-off-by: mrrr61 <mrrr61@outlook.com>

Author: mrrr3d

federated-learning-controller/examples/openfl/.gitignore

@@ -0,0 +1,6 @@
+.vscode
+
+data/
+.vscode
+.ipynb_checkpoints
+__pycache__

federated-learning-controller/examples/openfl/Makefile

@@ -0,0 +1,40 @@
+# Variables
+REGISTRY ?= quay.io/open-cluster-management
+IMAGE_TAG ?= latest
+DIST_DIR = dist
+APP = openfl-app
+PLATFORMS ?= linux/amd64,linux/arm64
+IMAGE_FULL_NAME = ${REGISTRY}/${APP}:${IMAGE_TAG}
+
+# Single-arch build (local)
+build-app-image:
+	openssl genrsa -out signer-key.pem -3 3072 && \
+	cd ${APP} && \
+	docker build -t ${IMAGE_FULL_NAME} . -f Dockerfile.workspace --secret id=signer-key,src=../signer-key.pem && \
+	cd ..
+
+# Single-arch push
+push-app-image:
+	docker push ${IMAGE_FULL_NAME}
+
+# Multi-arch build without push (export to local registry/engine not supported directly)
+build-platform-image:
+	cd ${APP} && \
+	docker buildx build \
+		--platform ${PLATFORMS} \
+		--tag ${IMAGE_FULL_NAME} \
+		--output=type=docker \
+		. -f Dockerfile && cd ..
+
+# Multi-arch push (requires --push during buildx)
+push-platform-image:
+	cd ${APP} && \
+	docker buildx build \
+		--platform ${PLATFORMS} \
+		--tag ${IMAGE_FULL_NAME} \
+		--push \
+		. -f Dockerfile && cd ..
+
+# Clean
+clean:
+	rm -rf $(DIST_DIR) __pycache__ *.spec

federated-learning-controller/examples/openfl/openfl-app/.workspace

@@ -0,0 +1,2 @@
+current_plan_name: default
+

federated-learning-controller/examples/openfl/openfl-app/Dockerfile.workspace

@@ -0,0 +1,35 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+# ------------------------------------
+# Gramine-ready Workspace Image
+# Usage:
+# $> docker build . -t openfl-workspace -f Dockerfile.workspace \
+#    [--build-arg WORKSPACE_PATH=/path/to/workspace] \
+#    [--secret id=signer-key,src=signer-key.pem]
+# ------------------------------------
+ARG BASE_IMAGE=ghcr.io/securefederatedai/openfl:latest
+FROM ${BASE_IMAGE}
+
+USER root
+SHELL ["/bin/bash", "-o", "pipefail", "-c"]
+
+WORKDIR /workspace
+COPY entrypoint.py /workspace
+COPY src/ /workspace/src/
+COPY plan/ /workspace/plan/
+COPY save/ /workspace/save/
+COPY cert/ /workspace/cert/
+COPY requirements.txt /workspace/
+RUN mkdir /workspace/logs && mkdir /workspace/data && mkdir /workspace/local_state
+
+RUN pip install --no-cache-dir -r ./requirements.txt
+
+# Build enclaves
+RUN --mount=type=secret,id=signer-key,dst=/key.pem \
+    cp -r /opt/venv/lib/python3.10/site-packages/openfl-docker/gramine_app/* /workspace/ && \
+    make SGX=1 SGX_SIGNER_KEY=/key.pem >> fx.mr_enclave && \
+    echo "$(cat fx.mr_enclave)" && \
+    chown -R user /workspace
+
+USER user
+ENTRYPOINT ["python", "/workspace/entrypoint.py"]

federated-learning-controller/examples/openfl/openfl-app/entrypoint.py

@@ -0,0 +1,120 @@
+import argparse
+import yaml
+from pathlib import Path
+import os
+import subprocess
+
+PLAN_FILE = Path("plan/plan.yaml")
+DATA_FILE = Path("plan/data.yaml")
+COLS_FILE = Path("plan/cols.yaml")
+
+
+def load_plan():
+    with open(PLAN_FILE, "r") as f:
+        return yaml.safe_load(f)
+
+
+def save_plan(config):
+    with open(PLAN_FILE, "w") as f:
+        yaml.dump(config, f, sort_keys=False)
+
+
+def update_server(args):
+    cfg = load_plan()
+
+    if args.cols:
+        if COLS_FILE.exists():
+            with open(COLS_FILE, "r") as f:
+                cols_config = yaml.safe_load(f) or {}
+        else:
+            cols_config = {}
+        
+        cols_list = [col.strip() for col in args.cols.split(',')]
+        cols_config['collaborators'] = cols_list
+
+        with open(COLS_FILE, "w") as f:
+            yaml.dump(cols_config, f, sort_keys=False)
+        print(f"[OK] Updated collaborator names in {COLS_FILE}")
+        cfg["data_loader"]["settings"]["collaborator_count"] = len(cols_list)
+
+    cfg["network"]["settings"]["use_tls"] = False
+
+    if args.server_ip:
+        cfg["network"]["settings"]["agg_addr"] = args.server_ip
+    
+    if args.server_port:
+        cfg["network"]["settings"]["agg_port"] = args.server_port
+
+    if args.num_rounds:
+        cfg["aggregator"]["settings"]["rounds_to_train"] = int(args.num_rounds)
+
+    if args.model_dir:
+        best_file = os.path.basename(cfg["aggregator"]["settings"]["best_state_path"])
+        last_file = os.path.basename(cfg["aggregator"]["settings"]["last_state_path"])
+        cfg["aggregator"]["settings"]["best_state_path"] = os.path.join(args.model_dir, best_file)
+        cfg["aggregator"]["settings"]["last_state_path"] = os.path.join(args.model_dir, last_file)
+
+    save_plan(cfg)
+    print(f"[OK] Updated server settings in {PLAN_FILE}")
+
+    os.execvp("fx", ["fx", "aggregator", "start"])
+
+
+def update_client(args):
+    with open(DATA_FILE, "a") as f:
+        f.write(f"{args.name},{args.data_path}\n")
+
+    cfg = load_plan()
+    cfg["network"]["settings"]["use_tls"] = False
+
+    if args.server_ip:
+        cfg["network"]["settings"]["agg_addr"] = args.server_ip
+
+    if args.server_port:
+        cfg["network"]["settings"]["agg_port"] = args.server_port
+
+    if args.num_rounds:
+        cfg["aggregator"]["settings"]["rounds_to_train"] = int(args.num_rounds)
+
+    if args.model_dir:
+        best_file = os.path.basename(cfg["aggregator"]["settings"]["best_state_path"])
+        last_file = os.path.basename(cfg["aggregator"]["settings"]["last_state_path"])
+        cfg["aggregator"]["settings"]["best_state_path"] = os.path.join(args.model_dir, best_file)
+        cfg["aggregator"]["settings"]["last_state_path"] = os.path.join(args.model_dir, last_file)
+
+    save_plan(cfg)
+    print(f"[OK] Updated client settings in {PLAN_FILE}")
+
+    os.execvp("fx", ["fx", "collaborator", "start", "-n", args.name])
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Update OpenFL plan.yaml")
+
+    subparsers = parser.add_subparsers(dest="role", required=True)
+
+    # docker run --rm image server --server-port 8080 --num-rounds 3 --cols client1,client2 --model-dir models
+    sp_server = subparsers.add_parser("server", help="Update server (aggregator) settings")
+    sp_server.add_argument("--server-ip", help="Aggregator IP address")
+    sp_server.add_argument("--server-port", type=int, help="Aggregator port")
+    sp_server.add_argument("--num-rounds", type=int, help="Number of rounds to train")
+    sp_server.add_argument("--cols", help="Comma-separated list of collaborator names for cols.yaml")
+    sp_server.add_argument("--model-dir", help="Directory for model files")
+    sp_server.set_defaults(func=update_server)
+
+    # docker run --rm image client --name client1 --data-path /data/client1 --server-ip 172.17.0.2 --server-port 8080 --num-rounds 3 --model-dir models
+    sp_client = subparsers.add_parser("client", help="Update client (collaborator) settings")
+    sp_client.add_argument("--name", required=True, help="Collaborator name")
+    sp_client.add_argument("--data-path", required=True, help="Path to collaborator data")
+    sp_client.add_argument("--server-ip", help="Aggregator IP address")
+    sp_client.add_argument("--server-port", type=int, help="Aggregator port")
+    sp_client.add_argument("--num-rounds", type=int, help="Number of rounds to train")
+    sp_client.add_argument("--model-dir", help="Directory for model files")
+    sp_client.set_defaults(func=update_client)
+
+    args = parser.parse_args()
+    args.func(args)
+
+
+if __name__ == "__main__":
+    main()

federated-learning-controller/examples/openfl/openfl-app/plan/cols.yaml

@@ -0,0 +1,5 @@
+# Copyright (C) 2020-2021 Intel Corporation
+# Licensed subject to the terms of the separately executed evaluation license agreement between Intel Corporation and you.
+
+collaborators:
+  

federated-learning-controller/examples/openfl/openfl-app/plan/data.yaml

@@ -0,0 +1,5 @@
+# Copyright (C) 2020-2021 Intel Corporation
+# Licensed subject to the terms of the separately executed evaluation license agreement between Intel Corporation and you.
+
+# collaborator_name,data_directory_path
+one,1

federated-learning-controller/examples/openfl/openfl-app/plan/defaults

@@ -0,0 +1,2 @@
+../../workspace/plan/defaults
+

federated-learning-controller/examples/openfl/openfl-app/plan/plan.yaml

@@ -0,0 +1,77 @@
+aggregator:
+  settings:
+    best_state_path: 6677/best.pbuf
+    db_store_rounds: 2
+    init_state_path: save/init.pbuf
+    last_state_path: 6677/last.pbuf
+    persist_checkpoint: true
+    persistent_db_path: local_state/tensor.db
+    rounds_to_train: 10
+  template: openfl.component.Aggregator
+assigner:
+  settings:
+    task_groups:
+    - name: learning
+      percentage: 1.0
+      tasks:
+      - aggregated_model_validation
+      - train
+      - locally_tuned_model_validation
+    - name: evaluation
+      percentage: 0
+      tasks:
+      - aggregated_model_validation
+  template: openfl.component.RandomGroupedAssigner
+collaborator:
+  settings:
+    db_store_rounds: 1
+    opt_treatment: RESET
+    use_delta_updates: false
+  template: openfl.component.Collaborator
+compression_pipeline:
+  settings: {}
+  template: openfl.pipelines.NoCompressionPipeline
+data_loader:
+  settings:
+    batch_size: 256
+    collaborator_count: 2
+    data_group_name: mnist
+  template: src.dataloader.KerasMNISTInMemory
+network:
+  settings:
+    agg_addr: 172.16.1.11
+    agg_port: 58112
+    cert_folder: cert
+    client_reconnect_interval: 5
+    enable_atomic_connections: false
+    hash_salt: auto
+    require_client_auth: true
+    transport_protocol: grpc
+    use_tls: false
+  template: openfl.federation.Network
+task_runner:
+  settings: {}
+  template: src.taskrunner.KerasCNN
+tasks:
+  aggregated_model_validation:
+    function: validate_task
+    kwargs:
+      apply: global
+      batch_size: 32
+      metrics:
+      - accuracy
+  locally_tuned_model_validation:
+    function: validate_task
+    kwargs:
+      apply: local
+      batch_size: 32
+      metrics:
+      - accuracy
+  settings: {}
+  train:
+    function: train_task
+    kwargs:
+      batch_size: 32
+      epochs: 1
+      metrics:
+      - loss

federated-learning-controller/examples/openfl/openfl-app/requirements.txt

@@ -0,0 +1,2 @@
+keras==3.11.3
+tensorflow==2.18.0